3262: Sports Commentary

2026-06-24T22:26:42Z

170.85.99.94: /* Explanation */

{{comic
| number = 3262
| date = June 22, 2026
| title = Sports Commentary
| image = sports_commentary_2x.png
| imagesize = 251x374px
| noexpand = true
| titletext = The plural of anecdote may not be data, but the singular of data is anecdote.
}}

==Explanation==
{{incomplete|This page was created at a statistically insignificant time, but it is the FIRST PAGE TO START WITH 3262. Don't remove this notice too soon. The title text's explanation should be expanded.}}

{{w|P-hacking}} is the academically problematic practice of attempting to come up with a question for which the data offers a significant ''p''-value (probability value), a subject [[882: Significant|previously covered]] in comic form. This is in contrast to correct scientific analysis, in which a realistic question is formulated clearly and then answered (or shown to be unjustified) with data. There are several issues with ''p''-hacking. One is that that larger data sets usually give more reliable results, so shrinking the data set indicates an effort to justify a conclusion, rather than a desire for accuracy. Another issue that the more different data sets you compare, the greater the odds of one of them showing a false correlation, simply due to statistical noise. An honest researcher would want to avoid such pitfalls, but someone trying to justify a conclusion might not care.

A method of ''p''-hacking involves analyzing subgroups to attempt to find significance when the full dataset does not yield statistically significant results; for instance, if a medical study didn't show an expected correlation, one might look only at data for male patients, and then only at male patients of certain age ranges, and so on, until they found a group that showed the desired correlation. Restricting data is warranted in some situations, but doing it to look for a particular result greatly increases the chances of misinterpreting statistical noise as a real result.

A similar effect is seen with sport commentators, and this is lampooned in the strip. Commentators often try to make predictions about developing situations by comparing them to past situations, such as previous competitions between the same teams. If commentators are trying to support a pet theory, however, they deliberately restrict themselves situations that ended in a particular way. By narrowing down the historical body with multiple qualifiers, they can justify a particular prediction. (A similar tactic was portrayed in [[2901: Geographic Qualifiers]])

Randall satirizes this with an example in which the restriction uses very specific criteria largely irrelevant to gameplay patterns in order to narrow down the subgroup sample size to a mere two games. The 0-2 record (there were two situations considered as comparable, and neither of them resulted in the result hoped for in this current case) reflects random noise much more than any significant insight. As well as being irrelevant to gameplay, their ''p''-hacking also makes the game sound like jargon, which can be confusing and difficult to understand. This is ironic given a sports commentator's job is supposed to be to explain the situation they are fronting, rather than making them more vague and incomprehensible. However, this may be the inevitable response to being left in front of the camera during breaks in play, or even during periods of gameplay that are nominally unremarkable — feeling the pressure to say ''something'', they will draw upon ever more obscure and irrelevant details to justify their (or their off-screen advisors') efforts and expertise to entertain and inform the viewing public.

The title text references an old saying in statistics: "The plural of anecdote is not data". This saying means a set of anecdotes do not constitute significant data, because anecdotes are heavily subject to selection bias, may be unreliable (as they're often not rigorously recorded or controlled) and usually don't come in large enough numbers to be significant. [[Randall]], however, argues that the reverse ''is'' true. By reducing the body of data to a single point (which is the ultimate extreme of ''p''-hacking), all you are left with is an anecdote, statistically worth nothing.

{{cot|A breakdown of the commentary's statement}}
The comparison being made is that "Over the last 36 years, they've gone 0 for 2 when they've scored in the 37th minute to lead 2-1 against a team whose country comes before theirs alphabetically." This contains the following basic stipulations:
;"Over the last 36 years, ...": Counting just the full FIFA World Cup competitions, assuming they qualified for every one, the total number of games that an international team will have played, prior to anything in this year's competition, would have been a minimum of twenty seven matches (i.e. playing the first round group-stages, playing once against each of the other three teams in their particular group of four). ''If'' they're ever successful enough in the group stage, they'd then progress through the knockout stages of the competition for as many matches as they avoid being knocked out plus one, and semi-finalists additionally get to play one more match to establish the third-place overall. On top of that, there are the various regional qualifying matches they will usually have had to play to even enter the main competition, plus any other international matches (e.g. '{{w|Exhibition game|friendlies}}' or other region-based inter-nation competitions) that may have been taken part in.
;"... when they've scored in the 37th minute...": A football game has a nominal 90 minutes of game-time, plus possible extra time. No team in the World Cup has scored any more than {{w|Hungary v El Salvador (1982 FIFA World Cup)|ten goals}} in a single game, but it is ''far'' more common for even winning teams to have scored just two or three times per game, statistically, the chances of scoring in any given minute is an insignificant detail. There is also effectively no useful analysis of a goal being in the 37th minute, as opposed to the 36th or 38th, and hardly any even in being between in the larger block between 30 and 40 minutes. The psychology of goal timings usually gravitates towards whether they were in the first or second ''half'' of the event (or, beyond that, in extra time), with most useful attention paid to those that occur right at the start of either half (one team immediately seizing the initiative on the field) or right at the end (when desperation, increased chance-taking or just player exhaustion can lead to much-needed/-feared game-changing goals once any attempt at mutually defensive play breaks down and possible goal-droughts are ended).
;"... to lead 2-1 ...":As an equivalent example, in the 2022 World Cup, 14 group stage games (out of 48) and 9 knockout stage games (out of 16) may have at some point reached a 2-1 scoreline for one or other team, depending upon the order the respective teams' goals occurred, making this a relatively rare situation to be in. For additional context, and most relevant to the full statement, that year's competition also saw just six group games that had scores that ''might'' have had a temporary 2-1 lead for the team that went on to lose, whereas ''no'' team with a 2-1 scoreline in the knockouts did not then go on to win that match.
;"... against a team whose country comes before theirs alphabetically.": In ''every'' international match (and others, excepting perhaps games used to train the team's players against each other), there will inevitably be one national team whose name is alphabetically prior that of their opponents, even if that features very similar names (such as a match between the two Koreas, using the most similar manner of naming, where {{w|North Korea national football team|Korea DPR}} would precede {{w|South Korea national football team|Korea Republic}}) and there would also be no clear reason why a naming issue (alone) would have any significant bearing upon match outcomes.
;"... they've gone 0 for 2 ...": (As the stated past consequence of all these specifically combined conditions.) Just ''two'' occasions satisfied all these conditions, out of possibly many tens of matches, and we are told that neither of them ended in a victory. Not only are the comic's precedents ''very'' rare, compared to all possible games (which, nevertheless seems to be even rarer in real life), but also this mini-'streak' of results is only a matter of history. In [[1122: Electoral Precedent]], increasingly convoluted situations may have previously been entirely predictive in possibly even several dozen instances... ''until they weren't''.
{{cob}}

This comic was published 11 days into the {{w|2026 FIFA World Cup}}. The World Cup was also the subject of [[3260: Messi]], published the previous Wednesday. Sports commentary was also the subject of [[904: Sports]].

==Transcript==
{{incomplete transcript|Don't remove this notice too soon.}}
:[Cueball and Ponytail are sitting at a table, looking at the wall behind them. On the wall is a screen showing a soccer field with some mostly unreadable score information above it. The only readable information is that the score is 2-1.]
:Cueball: They could be in trouble. Over the last 36 years, they've gone 0 for 2 when they've scored in the 37th minute to lead 2-1 against a team whose country comes before theirs alphabetically.
:[Caption below comic:]
:I wish sports commentators hadn't discovered p-hacking.

{{comic discussion}}<noinclude>

[[Category:Comics featuring Cueball]]
[[Category:Comics featuring Ponytail]]
[[Category:Soccer]]
[[Category:Statistics]]

explain xkcd - User contributions [en]

3262: Sports Commentary